Applied Digital Sensor Technology in the Analysis of Different Intensity Movements and Sensor Placements

全文

(1)National Taiwan Normal University Department of Physical Education PhD Dissertation. Applied Digital Sensor Technology in the Analysis of Different Intensity Movements and Sensor Placements. Student: Róbert János Füle Doctoral Supervisor: Tzyy-Yuang SHIANG, Ph.D. Republic of China, 103, July.

(2) Abstract Purpose: The study analyzed and compared movement modes and cycles, intensity levels and digital sensor positions. The target was to identify characteristics of body movements that could pave the way to a healthy and sustainable life. Revelations of the study provide potential information for creating a new sporting equipment and experience. Method: The observation of locomotion was executed with three high-tech Inertial Measurement Units (IMUs) that were attached to participants at three locations (shoe, wrist and waist). IMU was the fusion of a gyroscope and an accelerometer. Walk, Run and Jump movements were compared at two intensities. Result: The statistical analysis revealed an applicable correlation between movements and intensities. The simple effects test resulted in non-significant interaction between movements and intensities. This interaction served as a tool for comparing movement patterns with each other. Body movements included a series of gait cycles. The gait cycle was determined by acceleration data. Peak to peak intervals caused by the heel strike of the left foot were compared. Angular velocity data of gait cycles were benchmarked among different intensities. As a result the Shoe IMU measured the angular velocity on the frontal Y axis and discovered a regular sequence of plantar and dorsiflexion. Conclusion: Angular velocity data from the frontal axis clearly identified the movement features of walking, running and jumping. The acceleration data on the sagittal plane could distinguish between low and high intensity movements. The acceleration and gyroscope data determined the intensities and the body movements. The locomotion of lower extremities was widely explored. Waist and wrist IMU data even enabled the estimation of energy expenditure. Analysis methods of sensor signals were subject to investigation. Application of multiple digital sensors provided a unique opportunity for new observations.. Keywords: Movement, Intensity, IMU, Digital Sensor, Acceleration, Angular velocity. I.

(3) Acknowledgement I would like to express my most sincere appreciation to my supervisor, Distinguished Professor Tzyy-Yuang Shiang, who has given me extraordinary support throughout these years. Special thanks for granting me the opportunity to research in the biomechanics laboratory of the Department of Physical Education of the National Taiwan Normal University. Professor Tzyy-Yuang Shiang’s well established and esteemed position in both the biomechanics research and the related business community of Southeast Asia, Asia, the USA, the EU, and the whole world motivated me in my endeavor to complete the process of my PhD. Allow me to also express special appreciation to the chairman of the committee, Minister Chuan-Show Chen for giving his time to judge my work along with all distinguished members of the committee: Deputy-Minister Professor Michael Chi-Huang Huang, Professor Wei-Hua Ho, and Professor Chiang Liu. I am extremely grateful to all my colleagues, especially Peter Yin-Shin Lee, who is the Godfather of the biomechanics laboratory being aware of every single detail and part of it. Peter has been my mentor and friend for all these years. I am also very thankful to Dr. Shamu Chia-Hsiang Chen for creating such a good atmosphere and helping me with administrative tasks, which indeed provided me with an excellent working environment. I also have to state that I highly honor Yiping Chen’s support in all administrative challenges. With all this generous support of the Department of Physical Education I was provided with a platform where I could conclude my research in Taiwan. I have to express appreciation to my superior at work, the Representative of Hungary in Taiwan, Ambassador Levente Szekely and the First lady, Hedvig Szekely for understanding what I do and for caring about my studies. Most importantly I thank my wonderful wife, Szekely Zsuzsanna for always being there for me and patiently holding on until the research was completed. She has been taking good care of our children, Szilard Janos Fule and Greta Mei Fule and me with all her loving heart. I would also like to express my gratitude to my parents, Ilona Fule, and Janos Fule for so much that would be impossible to describe here with only a few lines, without them it would have been impossible to get here. I would like to thank Tamas Szombathelyi, Dr. Mark Felegyhazi, Laszlo Szilágyi and Laszlo Deli for being my true friend during my time at high school, in judo, and forever. Finally, I would like to thank Edgar Gonzalez for lecturing the English grammar in the text.. II.

(4) Table of Content ABSTRACT ................................................................................................................................ I ACKNOWLEDGEMENT ........................................................................................................ II TABLE OF CONTENTS ........................................................................................................III LIST OF FIGURES .................................................................................................................. IV LIST OF TABLES ..................................................................................................................... V CHAPTER I. INTRODUCTION ............................................................................................... 1 IMPORTANCE OF THE STUDY ................................................................................................................ 4 THE CURRENT STATE OF DIGITAL SENSOR APPLICATION IN PHYSICAL ACTIVITY AND IN HUMAN MOVEMENT STUDIES ........................................................................................................................... 5. DIFFERENT INTENSITY MOVEMENTS AND SENSOR PLACEMENT ......................................................... 9 PROPOSED APPROACH TO SOLVE THE RESEARCH PROBLEM............................................................. 17. CHAPTER II. METHOD ......................................................................................................... 20 PARTICIPANTS .................................................................................................................................... 20 EQUIPMENT........................................................................................................................................ 20 PROTOCOL ......................................................................................................................................... 23 DATA PROCESSING ............................................................................................................................. 28 STATISTCAL METHOD ........................................................................................................................ 29. CHAPTER III. RESULTS ........................................................................................................ 30 CHAPTER IV. DISCUSSION ................................................................................................. 45 CHAPTER V. CONCLUSION……………………………………………………………….51 REFERENCE ........................................................................................................................... 53. III.

(5) List of Figures FIGURE 1 Treadmill ................................................................................................ .21 FIGURE 2 The IMU sensor ....................................................................................... 22 FIGURE 3 The Software of the IMU ........................................................................ 22 FIGURE 4 Protocol of the Main-test ......................................................................... 25 FIGURE 5 Wrist, pelvis, and shoe locations of combined sensors ........................... 26 FIGURE 6 Three types of different body movements and two intensities................ 27 FIGURE 7 The result of acceleration (left) and angular velocity (right) in X and Y axis respectively ................................................................................................ 33 FIGURE 8 Definition of one gait cycle at Slow walk from the shoe IMU ............... 34 FIGURE 9 Data of Slow walk from six axes (gyro on the left, accelerometer on the right)…………………………………………………………………………...34 FIGURE 10 Comparison of Slow walk and Fast walk intervals ............................... 35 FIGURE 11 Comparison of the Slow run and the Fast run ....................................... 36 FIGURE 12 Shoe IMU data at the Slow walk........................................................... 36 FIGURE 13 Shoe IMU data at Slow run ................................................................... 37 FIGURE 14 Waist IMU data at Fast walk ................................................................. 37 FIGURE 15 The waist IMU patterns at Fast run ....................................................... 38 FIGURE 16 The wrist IMU patterns at Slow run ...................................................... 38 FIGURE 17 Comparison of the Low(left) and High(right) intensity jump ............... 39 FIGURE 18 Wrist and shoe(foot) gyro data of the High intensity jump ................... 40 FIGURE 19 Wrist and shoe(foot) gyro data of the Low intensity jump…………....41 FIGURE 20 Comparison of the IMU acceleration patterns of the shoe(up) and waist(down) ....................................................................................................... 42 FIGURE 21 Low(left) and High(right) intensity Walk, Run, and Jump movements.43. IV.

(6) List of Tables TABLE 1 Results of interaction effects measured by the 3 IMUs….……………...31. V.

(7) CHAPTER I. Introduction Regular physical exercise is one of the essential elements of maintaining good health. An active lifestyle and even moderate intensity leisure time activities can reduce the risk of mortality (Bucksch, 2005). Many people do no sports at all or do not move enough to burn at least the amount of energy that is taken in every day. Increased energy intake over energy expenditure on the long run might lead to a sustained positive energy balance, which is one of the most important causes of obesity (Sazanov et al., 2011). In order to avoid becoming overweight and to remain healthy, physical activity (PA) must be increased to counterbalance the calories consumed. To know how much activity is sufficient for the counterbalancing, the exerted energy needs to be monitored at all times. Today wearable motion sensors are available for monitoring purposes. With the general improvement of living conditions, people have a higher life expectancy rate. People need to stay fit to prevent health problems and the elderly need to remain healthy in order to sustain an independent lifestyle. To support the care of the elderly, and the chronically ill, or the ones with special needs, researchers have to work out scientific solutions to create systems that can monitor the PA. Such systems are available like the Human Activity Recognition (HAR) (Ugulino et al., 2012). HAR is a non-static method of data collection. It gathers crucial information which can help with changing people’s physical condition. Its measuring unit can be attached to users and freely carried anywhere. Publications on small sized specialized measuring units attached to participants have shown an increasing number due to the growing demand of research driven by the expanding market of wearable devices that monitor and measure PA. Health oriented companies continuously develop the technology of micro-electro-mechanical system, MEMS, which is a term by Prof. R. Howe, 1989 (Chollet & Liu, 2008). MEMS based devices make new ways of non-static observations possible by micro fabricating small portable devices that can measure the amount of PA of various movements. These PA monitors 1.

(8) can support the prevention of obesity, enhance the performance of training, or be applied in rehabilitation of clinical patients. The most commonly used sensors for body movement detection can be divided into three major categories: gyroscopes accelerometers and pedometers. Gyroscopes are angular velocity sensors. MEMS gyroscopes were once developed for the military in the field of missile navigation. Later the technology was adopted by the automotive and the electronic sectors. Gyroscopes are also applied in motion capture studies and often in combination with accelerometers. The accelerometer after revolutionizing the automotive and the computer industry became a common measuring equipment of joint kinematics studies (Fong & Chan, 2010). These portable accelerometers can facilitate extended data collection. In PA studies accelerometers overcame experimental limitations (Butte et al., 2012, Chen et al., 2012, Reilly et al., 2008) due to being small, light, and mobile. Accelerometers also provide a noninvasive way to measure the frequency, duration, and intensity of PA. The information provided by accelerometers can indicate energy expenditure (EE) (Butte et al., 2012). Accelerometers can measure the daily EE from physical exercise or from non-exercise activity thermogenesis (NEAT), which means the time we spend on posture allocations like sitting down, standing up, and lying down (Sazanov et al., 2011). Accelerometers can also detect falling and they are applied for monitoring elders at the risk of falling (Senden et al., 2012), or other multiple purposes as for gait analysis (Takeda et al., 2009). In human research pedometers, counting steps may have been the first to record PA (Mehl & Conner, 2012). They are light, mobile, noninvasive, and reasonably priced equipment to monitor PA. According to Google trend most people use pedometers in the USA, Canada, Australia, and in the UK. The most users are in the US, where pedometers are so popular that even movements were initiated such as the Quantified Self Movement, in 2007. Users of pedometers aim to get a comprehensive picture on their activities, and mental health. To have data on certain basic activities can help in 2.

(9) managing resources much better, for example quantifying the daily routine movements like walking, cooking and shopping can help with the creation of new sporting experiences that burn more calories. Pedometers can assess PA both in laboratory and in living conditions. However the accuracy and reliability of pedometers must be validated, because both of these key features tend to be affected by different intensity movements and different environments. The cutting edge technology of MEMS has advanced rapidly, which resulted in smaller sized sensors, more precise operation, and upgraded designs. Today people who would like to be more aware of their energy output can simply wear MEMS sensors. These light weight mobile units have become the major tool of data collection in PA and human movement studies. Sensors can operate almost anywhere therefore human motion experiments are not strictly restricted to the laboratory environment. Researchers can easily apply them for tracking a wide range of activities. Sensors are now capable of gathering data on the physiology of the heart rate, body temperature, heat flux, acceleration, and angular velocity. These small devices can easily be carried by athletes of almost any sport, health oriented people, and patients of rehabilitation programs, which makes observations possible almost anywhere indoors, outdoors, at home and at the office. Marker sets are not mandatory, signals can be continuously captured and trajectories are not lost due to blocked markers. The latest digital devices are connected to the internet with Global Positioning System (GPS), which enables the recording of speed and displacement almost everywhere in real life situations. This satellite-based system provides information on the overall environment with geographic data. Furthermore the data processing and analysis of the GPS is a rather complex task. In addition the GPS observation is limited to outdoors. However the application of MEMS sensors in free-living conditions involves limitations in the identification of different movement types and the level of movement intensity. Therefore this study applied 3 Inertial Measurement Units (IMU) to observe different intensity movements in a laboratory setting. IMU placement has also been investigated. IMUs were located at the center of mass and on the limbs to capture 3 types of body movements. 3.

(10) Importance of the Study. There is an abundant amount of studies about PA recognition. In these researches only a very little emphasis has been put on the movement analysis. There is a major difference between a complete PA and one single locomotion cycle. In general each PA consists of numerous movement cycles. One movement cycle lasts for seconds or even less. The movement cycle repeats itself a hundred or a thousand times within one single PA. Analyzing repeated movement cycles will create data that can identify ranges and tendencies within a certain PA. These features are crucial in determining the intensity level of movements. The importance of gathering information about the intensity of the exercise, and the energy expenditure during exercise is high. It is essential to quantify PA in order to upkeep a high quality healthy and sustainable life. It is the reason why this compact technology is applied not only in professional sports but in the field of health care. The key concept of these small sized intelligent data recording MEMS devices is to capture and quantify all movements including steps taken, stairs climbed and even the posture changes. These real time figures based on scientific knowhow can support people with drawing essential conclusions on their daily routine activities and can also give invaluable advice on how to change individual movement habits for a better life. This study endeavored to collect comprehensive data with MEMS sensors. Information gathering went beyond the possible application of various measurement units in recent PA studies. MEMS sensors may track walking, jogging and even the footwork of different sport moves in order to better understand and improve the moves by shortening their execution time or by making them more efficient. Features of each sensor position need to be investigated because pedometers, gyroscopes, and accelerometers placed on different locations on the body may alter the overall outcome of an experiment. However movement data may be influenced by many other factors beyond sensor placement. Different body postures and movement might require 4.

(11) specific sensor adjustments. This study applied high-tech MEMS devices in order to contribute to the knowhow on the positive effects of different duration and intensity movement on human health.. The Current State of Digital Sensor Application in Physical Activity and in Human Movement Studies. Digital sensor application for monitoring PA became highly emphasized in the last few years. Kang et al. (2009) attempted to provide a comprehensive analysis on the effectiveness of interventions using pedometer as a motivational tool. In applied pedometer studies there are certain variables that might influence the effectiveness of interventions (age, sex, length of intervention). In this study between 2000-2007 103 theses, dissertations, and articles were cross checked on Medline, Pubmed, Sportdiscus, Google Scholar, AAHPERD National Convention and Exposition database, and Proquest. Kang et al. (2009) used a statistical system to compute all data and concluded that the application of a digital sensor capable of quantifying movements has a positive effect on the increase of PA in intervention studies. Effect size (ES) was calculated based on the variables such as age, sex, length of intervention. Variables affected the magnitude of PA increase. The ES was much higher at the female group interventions compared to the male interventions. The ES was only slightly higher due to the 15 week long interventions than due to the 8 week long interventions. The increase of PA was already significant at 8 week long interventions. Thus the application of a digital sensor for motivating people can be successful even within a relatively short period of time. The ES was the highest when intervention strategy of 10,000 steps per day was set as a goal. This type of intervention was only applied for adults therefore it might not be effective for elder persons or children. This study proved that pedometers increase motivation and are effective in increasing PA. However this study did not differentiate between 5.

(12) movements or intensity levels. Inclusion of different intensity movements such as running as a variable could have been a more comprehensive method to calculate the ES. Many PA studies have focused on gait analysis and tried to categorize the changes in static postures (Czabke et al., 2011, Kavanagh & Menz 2008, Kwapisz et al., 2011, Sazonov et al., 2011). In these researches different locomotion and average daily life movements were observed. However these studies were limited because of an incomplete wearable sensing system. Adding an accelerometer would have made the measurement of movement intensity possible. Single location MEMS sensors can be built into shoes. Sazonov et al. (2011) revealed that the single shoe based approach (with one combined acceleration and pressure sensor) matches or at certain measurements such as accurate recognition of postures and motions (ascending and descending stairs) may even outperform other single-location methodologies. The single shoe sensor outperformed ankle and wrist based single sensors. The shoe based accelerometer reached an accuracy of 95% in posture and activity recognition. However its limitation compared to other type of sensors was that it could measure data only when the shoes were on. Another challenge regarding the application of this study is that it could be more suitable for monitoring office or school activities since all of its data were captured in laboratory conditions where the acceleration and pressure could have been different from the outdoor conditions. Application of a single measurement unit may delimit observation. Application of multiple digital sensors for data collection on different human moves and body postures may have provided more data, thus enabling a more precise PA observation. In other single sensor human movement studies like in the study of Lee et al. (2009) one accelerometer was applied on the left waist of 5 participants. This study successfully reached high accuracy in recognizing standing, sitting, walking, lying, and running postures (99.5%). A specifically designed accelerometer was used for the research. The greatest asset of this accelerometer set was its accuracy and individual design. It was not a widely available commercial device unlike the one used in the 6.

(13) study by Kwapisz et al. (2010). This research concluded an experiment on phonebased accelerometers for activity recognition. A popular cell phone was used to collect data, which was simply positioned inside the 29 participants’ pocket. The tri-axial accelerometer of the cell phone was the main tool of data collection in this experiment. The smart phone had various functions encompassing GPS, camera, microphone, light sensor, and a temperature-and direction sensor. The mobile phone similarly to the ordinary GPS could detect the direction of our planet’s gravity so conclusions were drawn regarding the postures of participants. Although the activity recognition was relatively high (90%) with a single smart sensor, the study did not provide real time results because the phone was not directly connected to the activity recognition model. With an additional on-line MEMS device this study could have provided more real time results. It is rather difficult to predict the one single location on a participant’s body which will provide relevant features for a certain group of movements, therefore applying accelerometer alone as the single data source of a study can be a limitation. It is a quite challenging task to pick the right tool for measuring human movements. In the study of Mayagoitia et al. (2002) a gyroscope complemented the accelerometer in one sensor unit. It enabled the calculation of movement angles. In the absence of movement the accelerometer unlike the gyroscope can measure gravitational acceleration of a certain body segment (Aminian, 2006). However accelerometer may not measure the difference between walking at a given speed carrying on different loads of weight. In order to monitor movement angles Mayagoitia at el. (2002) combined the gyroscope with the accelerometer. With the support of this merged sensor the energy output was successfully measured. The accelerometer data showed deformed peak values because sensors resonated during heel strike. However the gyroscope was able to show the orientation of moves without interference by the same heel strike. The accelerometer and the gyroscope complemented each other. The accelerometer measured angular velocity and the gyroscope estimated rotation. Accelerometers could capture body segment inclination when resting unlike the gyroscope. Gyroscopes were not sensitive to gravitational acceleration unlike accelerometers. Therefore the combination of the accelerometer 7.

(14) with the gyroscope into one integrated measurement unit provided a more comprehensive sensor set for data collection. Such a fusion system was also applied in health care to measure the effectiveness of active physiotherapy or to remotely observe the elderly at the risk of falling. It could intervene in case of an emergency by sending an alarm signal to a third party. It could also be used to perfect the movements of athletes in professional sports (Varkey et al. 2011). Further applications of this sensor fusion could involve the estimation of energy output to help people in weight control. Research has used the characteristics of the accelerometer and gyro signals generated during locomotion to investigate the relationship between PA and energy expenditure. These studies involved preset locomotion and intensities to verify the results of acceleration signals and oxygen intake. As a result the accelerometer and the gyro became an adopted measurement tool to replace oxygen analysis when assessing energy expenditure and determining PA by cut points (Aadland and SteeneJohannessen, 2012, Bao and Intille 2004, Kurihara et al., 2012, Takeda et al., 2009, Zhang et al., 2012). Defining cut points with such methods is challenging because it requires post-processing and complex calculations such as Fourier transform or Wavelet analysis. Sensors such as gyroscope and accelerometer fixed at multiple locations and the combination of various measurement units provided the option of a more thorough collection of movement data. Györbíró et al. (2009) applied “Motion Bands”, combined sensors where each unit had a tri-axial accelerometer, a magnetometer, and a gyroscope. This study similarly to the study of Liu et al. (2012) successfully applied more monitoring devices combined into one unit placed on multiple locations of the human body to assess PA. The study categorized 13 activity types and measured related energy expenditure. The recognition accuracy was 12.3% higher with the sensor fusion technique, than with the single hip accelerometer. Furthermore with the sensor fusion technique energy expenditure was predicted with a 22.2% lower rate of statistical error than with the single sensor on the hip. Results support that combined units or sensor fusion techniques are more effective in differentiating between activity types and assessing energy expenditure than single accelerometer methods. This may 8.

(15) indicate that even if single sensors are placed at multiple locations, perhaps they would not be as effective as combined sensor units at multiple locations. Often sensors were fixed at multiple locations to monitor a specific group of body movements with a specific target. The angular velocity of foot pronation during run movement was measured by Shih et al. (2013). The study applied gyro sensors to predict the fatigue status of participants’ lower extremities. There were numerous studies that measured the risk of falling among senior citizens or monitored how much activity was delivered by clinical patients of Parkinson’s disease (Senden et al., 2012, Zhang et al., 2011). Although these studies had their specific target they were delimited by the challenge of the complex calculations and post processing protocols. Electronic sensors have replaced mechanical pedometers when assessing PA. The high-tech MEMS sensors are accurate and have functions similar to complex physiological instruments (Hendelman et al., 2000, Welk et al., 2004). However the fact that movement analysis with MEMS sensors still has its limitations remains. Different movements require different identification approaches (Brandes et al., 2006) and complex sometimes complicated calculation techniques (Favre et al., 2008, Liu et al., 2009).. Different Intensity Movement and Sensor Placement. Atallah et al. (2011) applied combined measurement units including a 3D accelerometer at multiple locations of participants’ body and grouped every day activities into 5 intensities as follows: 1. Very low level activity ergo lying down; 2. Low level activities: Preparing food, Eating, Drinking, Reading, and Getting dressed; 3. Medium level activities: Walking in a corridor, and Treadmill Walking at 2 km/h, Vacuuming, and Wiping tables; 4. High level activities such as Running in a corridor, Treadmill running at 7 km/h, and Cycling; 5. Transitional activities like Sitting down 9.

(16) and getting up or Lying down and getting up. The study successfully identified the optimal positions for placing sensors. The wrist sensor trended to show significant results for the Very low level activities and the Medium level activities such as walking at 2 km/h; The waist sensor collected significant results for the Low level activities, and one of the Transitional activities the Sit down and stand up posture change; The Knee and the ear-worn sensor provided significant figures for the High level activities like Running in corridor, Treadmill running at 7 km/h, and Cycling. Sensors placed at multiple locations on the body may have an increased chance of detecting more activity types compared to the single sensor solutions. Furthermore the knee position was proven to be an optimal location for measuring all intensity movements. The high intensity movements were captured the best with the knee and the head sensors. This study did a comprehensive analysis on different intensity movements. Limitation of this study was the definition of 2 km/h walking as medium level PA, and using the 7 km/h treadmill running as high intensity movement. Seven km/h is approximately 1.94 m/s, which is a velocity relatively closer to the bottom of the threshold of the average Preferred Transition Speed, PTS (Hreljac, 1995). A faster than 2 km/h walking speed and a different type of movement at the same 7 km/h speed would have been a better representative of middle and high intensity movements respectively. The energetically optimal transition speed, EOTS occurred at a significantly higher speed than PTS (Rotstein et al., 2005, Tseh et al., 2002). EOTS=2.24m/s compared to PTS=2.06 m/s. The energy expenditure of EOTS was significantly higher than at PTS. Therefore the EOTS demonstrates the high intensity movements more than PTS. However Hreljac’s (1995) study scientifically established a theory by kinematic variables that in the human movement the main reason for shifting into running from walking is to prevent overextension of the ankle muscles responsible for dorsiflexion in the gait cycle. These rather small ankle muscles produced burst values of EMG and produced a significant amount of energy starting from the toe-off phase until almost the end of the swing phase. Therefore PTS, 1.9-2.1 m/s may be considered as a velocity suitable for measuring high intensity movements. Hreljac’s (1995) study was limited because the use of EMG would not have been ethical on human subjects therefore the research had to be run with horses. Deeper 10.

(17) analysis of PTS would be only possible on data deriving from human participants. Researchers tried to predict human PTS from the moment when participants were not capable of maintain walking and simply started running (EOTS). PTS was reached before EOTS based on an early study by Diedrich and Warren Jr. (1995). The range of PTS varied between 2.07-2.2 m/s. Data of the study indicated that the walk-run transition trended to occur at a speed of 2.1 m/s to minimize energy costs. The same finding was supported by Rotstein et al. (2005), where PTS typically happened at the speed of 2 m/s. Although in Hreljac’s (1995) study participants were not likely to shift between walking and running when it was energetically ideal. It may indicate that EOTS requires more energy and the intensity level is relatively higher at EOTS than at PTS. Therefore PTS during different types of movements could be subject to investigation in order to classify the exact range of PTS and assess energy expenditure related to this velocity. A complex measurement platform consisting of various monitoring equipment could be applied to measure the intensity and the energy expenditure of different movements at PTS. It is a rather challenging task to separate movement intensities and to define cut-points. However recommendations do exist on best practices of setting up the cutpoints that distinguish between moderate-to-vigorous physical activities, MVPA. Cutpoints can be defined with calibration studies where participants perform different intensity movements such as running, or walking and where accelerometers measure movements. With accelerometers for example indirect calorimetry can be performed to assess energy expenditure. Physical activities are usually divided into movement cycles and by quantifying these cycles the metabolic equivalents, METs can be indirectly calculated. Thus movements may be categorized by MVPA. However after the accuracy of cut points is validated by an independent validation study a criterion measure can be applied to compare whether the cut-points that estimate MVPA are close to the that of a criterion measure. Despite the considerable amount of researches in biomechanics on cut points there is still no consensus on where exactly moderate physical activities end and from where movement intensity could be counted as vigorous. The existing calibration and validation studies need to be reviewed. Kim et 11.

(18) al. (2012) reviewed a total number of 278 relevant articles that were filtered and 12 calibration and eventually 4 validation studies were compared. One of the major filter criterions was that studies had to apply Actigraph accelerometer. Other criterions were that the sample age range was 18 years or younger and that 50 % of these studies were concluded in free living conditions. The number and the length of movement cycles for cut points, the protocol, the demographic information of participants and the analytical procedures were compared. Calibration studies were scored based on their compliance with the criteria published by Feedson et al. (2005) and Welk et al. (2005). The criteria were a minimum of 10 participants; application of a wide variety of physical activities; the length of movement had to be less than 60 seconds and if it was less than that the study was rated with higher value. The fourth criterion was the use of a biological standard such as the indirect calorimetry, which provides quantifiable accelerometer data on movement cycles. When slow walking, slow running and fast running movement cycles were benchmarked, the results showed that the estimated mean energy expenditure equation precisely predicted the step count. For example at the velocity of 1.6 km/h walking on treadmill the measured energy expenditure and the predicted energy expenditure with calorimetry was in compliance with each other. However in the study by Wickel et al. (2007) the 3.2 km/h, 4 km/h, 4.8 km/h, and the 6.4 km/h treadmill walk and run showed a significant difference between measured and predicted values. Trost et al. (2006) used a 3 METs threshold for MVPA in the study of participants consisting of 45 boys and girls. Four different intensity movements were performed the normal walking, brisk walking, easy walking, and the easy and fast running. Movement cycles were 5 seconds. None of the equations used for indirect calorimetry estimated precisely the energy expenditure among the different intensity movements with the exception of the slow running. Kim et al. (2012) proved that Evenson cut points provide a significantly high precision when estimated energy expenditure and measured results are compared in the age group between 5-15. In Kim et al. (2012) study basketball, slow walk, brisk walk, and treadmill walk movements were compared. The agreement between Evenson et al. (2008) and Freedson et al. (2005) results were significantly high. However results indicate that only few studies were capable of precisely estimating energy expenditure with indirect calorimetry 12.

(19) (Trost et al., 2006, Feedson et al., 2005, Welk et al. 2005, Evenson et al., 2008). Therefore it is too premature to set up universal cut-points that would identify intensity levels (Kim et al., 2012). Further independent research could reveal the cut-points in PA. Although Actigraph accelerometers were able to collect accurate data on the low intensity slow movement (Feedson et al., 2005, Welk et al., 2005, Trost et al., 2006), the slow velocity trended to challenge the pedometer technology. Pedometers at low intensity could not accurately detect movements (Ayabe et al., 2010). In the study of Ayabe et al. (2010) two types of pedometers were applied and compared. The Yamasa pedometer below the speed of approx. 1.1 m/s significantly underestimated the number of steps while the Lifecorder did not. As a consequence to that, the measurement error for the number of steps during a treadmill walking validation exercise was much higher with the Yamasa sensor than with the Lifecorder sensor. The Lifecorder reached a significantly lower measurement error at the speed of approx. 0.9 m/s. Both sensors at a slower than approx. 0.9 m/s speed showed significantly different values compared to the actual value. After the in-lab treadmill validation the research was conducted in live conditions. There were 28 elder and 17 younger participants. Elder participants mostly executed light-intensity PA such as slow walking, which provided less than 3 metabolic equivalent (METs) energy output. The slow walking speed was less than approx. 0.9 m/s. Results from both pedometers were not in line with the actual values. Furthermore the difference between the values of the two pedometers was the most significant below the speed of approx. 0.9 m/s. Therefore the study suggested that people using pedometer for walking at low intensity with slow speed must confirm the accuracy of the pedometer to be able to properly quantify the daily PA. In the study by Crouter et al. (2003) 10 different pedometer brands were applied. Only 6 of them set the maximum error threshold of miscounting steps as accurate as 3%. Five out of these 6 sensors were calibrated according to Japanese industrial standards and there was one Taiwanese with the same accuracy. Only four out of ten devices were able to gather valid information from slow walk at the speed not more than 0.9 m/s. Pedometers used for measuring slow walk below the speed of 13.

(20) 1.1 m/s might be less accurate. However there were pedometers that could record valid data even at as slow speed as 0.9 m/s Crouter et al. (2003). The application of MEMS devices for the purpose of measuring low intensity movements has to be thoroughly evaluated, because the inaccuracy of pedometers may affect the overall outcome of the study. Above 4 km/h (approx.1.1 m/s) a more “robust criterion measure” such as the accelerometer may be added to strengthen results of studies related to footwork analysis (Cocker et al., 2012). Therefore measurement units require careful selection, especially at low intensity movements because errors regarding the over-or underestimation of steps might occur during the recording process. Sensor position is one major area of analysis. A study by Ugulino et al. (2012) collected 144 Institute of Electrical and Electronics Engineers (IEEE) articles related to human activities and body posture with application of accelerometers. They had come to the conclusion based on a comprehensive review of 69 scientific IEEE papers in the related field that wearable accelerometers are most commonly positioned close to the center of mass on the waist and on the chest. The most valuable data was derived from these sensor locations. Other accelerometer placements were listed ranging from the waist, left thigh, and right ankle to the right arm. Accelerometers are highly suitable to measure human movements in free living conditions. Therefore a further study of the sensors located on the waist could validate the previous results on MEMS accuracy. Data gathered from waist sensors play an important role in studies related to falling. In the 8 month study of Kangas (2011) sensors were attached to participants’ wrist, waist and forehead. The study was concluded in two countries and few real falls were detected amongst which there was an actual hip fracture. The accelerometers fixed on the waist trended to produce the most significant results at falls and proved to be the most reliable sensor placement location for discriminating falls from activities of daily living.. 14.

(21) Györbíró et al. (2009) attached sensors on participants’ wrist, hip, and ankle. The study was able to successfully distinguish 6 different types of motion patterns. Data was transmitted to a smart phone carried by participants. Mannini and Sabitini (2010) applied 5 accelerometers on participants’ hip, wrist, thigh, ankle, and arm recognizing 20 different moves. The maximum range of acceleration was reached during running at the ankle position. The ankle accelerometers trended to provide the most significant data when fast walking and running moves were observed. Acceleration during PA differed depending on the location of the sensor on the body. Acceleration figures showed an increasing tendency from sensors fixed on the hip to sensors fixed on the ankle. Tapia et al. (2007) created a real-time system recognizing a wide range of 30 different moves with 5 accelerometers and achieved significantly high activity recognition (94.6%). In this study accelerometers were placed on participants’ wrist, ankle, arm, thigh, hip, and there was a heart rate monitor on their chest. Sensors placed at the wrist and ankle enabled the accurate recognition of numerous moves and body postures. Therefore application of multiple measurement units may give more chance for capturing relevant results of certain human movements. Although when multiple sensors are applied in real life conditions, researchers always have to consider the actual future function of the sensor set in order to avoid causing inconvenience for users. Balance needs to be established between very high activity recognition and many sensors versus a more lifelike application of measurement devices with less sensors that can still maintain a reasonably high activity recognition rate. In order to reduce the number of MEMS in one study more ideal locations need to be further explored. In a study by Ermes et al. (2008) accelerometers combined with GPS were applied on the wrist and the hip. In earlier studies the wrist, hip, and ankle sensor positions have also trended to be highly useful in the reliable and accurate measurement of various movements (Györbíró et al., 2009, Tapia et al., 2007). In the 15.

(22) study of Ermes et al. (2008) the wrist and the hip sensors achieved 89% of recognition rate successfully identifying 9 different activities. Liu et al. (2012) fixed vector machines on the wrist, hip, and abdomen of a considerably sized 50 participant population. This study indicated that when sensors combining multiple measuring units were fixed on participants’ wrist, hip, and abdomen the energy expenditure and the activity types were identified more accurately. Atallah et al. (2011) found that wrist accelerometers were the most effective in measuring treadmill walking at 2 km/h, or walking in a corridor while waist sensors were the most effective in collecting data on “Low level activities” and vertical movements like sitting down and getting up 5 times or lying down and getting up 5 times. The ear worn sensor caught the treadmill running at 7 km/h the most accurately. This sensor captured the change of body posture between running and walking. Wrist and waist locations for sensor placement trended to provide comprehensive sensor coverage for walking, and vertical moves respectively. However the setting of walking intensity at 2 km/h might be a limitation to this study, because the slower than 4 km/h walking speed influenced the data capturing ability (Cocker et al., 2012). In the study by Holbrook et al. (2009) HJ-720 pedometers placed in participants’ backpack showed 3.3-3.5% of absolute percentage error (APE). APE is originally calculated from the under-or overestimation of steps by a pedometer. The same sensors placed in participants’ pants pocket produced a lower 2-2.9% of APE. Hasson et al. (2009) in the validity study of HJ-112 pedometer during treadmill walking observed 4 different sensor positions the hip, pants pocket, chest shirt pocket, and neck. This study found that the sensor fixed on the hip produced the smallest random error of 1.2% and pedometers in the pants pocket had only little effect on validity. Neck pedometers had a smaller error rate than the pedometers in the pants pocket. In the study by Cocker et al. (2012) neck pedometers type HJ-203, similarly to the results of Hasson et al. (2009), showed the most acceptable APE compared to pedometers in pants pocket, and in carrier bag during the controlled test. In this study 16.

(23) 40 people participated in the controlled test and 54 in the free living condition test. The free living condition test did not confirm the results of the controlled test regarding the neck pedometers. In the free living conditions test there was not any statistical difference between the pants pocket and the neck pedometers. Pedometers fixed on the hip might have a decreased accuracy rate due to participants’ abdominal fat, which hypothesis Holbrook et al. (2009) did not concur. To conclude pedometers at low intensity PA may fail to provide reliable data on human movements (Cocker et al., 2012). Pedometers located in the pants pocket or bag positions resulted in high error percentage during data collection, and it delimited the accuracy of the PA observation (Ayabe et al., 2010, Holbrook et al., 2009). Accelerometers especially when fixed at a relatively high point on the body were less effective in assessing body movements (Atallah et al., 2011). However the waist, hip, ankle, and wrist accelerometers trended to be ideally accurate locations when monitoring PA (Ugulino et al., 2012, Mannini and Sabitini 2010, Ermes et al., 2008, Liu et al., 2012).. Proposed Approach to Solve the Research Problem. Findings related to the application of sensors in PA and movement studies are increasing rapidly by the continuous advancement of MEMS technology. At present the sport science research includes the application of MEMS sensors and the integration of such portable devices into extended data collection on PA. However many of these studies fail to break down PA into movement cycles in the analysis of the intensity of human movements (Kang et al., 2009, Czabke et al., 2011, Kavanagh & Menz 2008, Kwapisz et al., 2011, Sazonov et al., 2011). MEMS sensors can pick up and identify the features of movements in order to categorize body movements not only by type, but by intensity. Furthermore researchers find it difficult to select the most effective methods for movement identification, especially when analyzing a 17.

(24) considerable amount of data. Accelerometers detect speed, while gyroscopes detect rotation and orientation of human joints. Measurement of acceleration is an economic and accurate solution for estimating physical activity. In the study by Shiang et al. (2012) accelerometers were applied to capture different movements. High correlation was discovered between the heart rate, the running speed and the acceleration data. Single sensor studies limited data capturing in human motion studies by missing the relevant location, which could identify a certain group of movements. Many pedometers failed to provide reliable data on low intensity human movements (Ayabe et al., 2010). It is rather difficult to find the ideal location on the human body for fixing pedometers. Moreover pedometers at certain locations have a high measurement error rate and may under or overestimate the number of steps (Holbrook et al., 2009, Cocker et al., 2012). Therefore pedometers were not applied in this study. In previous studies monitoring different intensity movements with the combination the accelerometer and the gyro sensor offered an effective unit fusion in the data collecting platform (Varkey et al., 2011, Mayagoitia et al., 2002, Aadland and Steene-Johannessen, 2012, Bao and Intille 2004, Kurihara et al., 2012, Takeda et al., 2009, Zhang et al., 2012). Walking, running, and jumping as variables are important so there is yet to investigate the effect of these movement types on MEMS data. Walking and running are basic movements equally important for athletes and for people in general. Jumping beyond being a common movement is an essential skill for athletes. Jumping ability translates to numerous sports where strength, speed and flexibility are a must. Furthermore the observation and improvement of the vertical jump is also crucial to elite athletes of many Olympic and audience sports such as basketball, soccer, handball, or volleyball. Therefore in this study data and signals were collected by a combined Inertial Measurement Unit IMU sensor, the fusion of an accelerometer and a gyroscope. IMU sensors were placed at multiple locations where previous studies trended to discover the most significant results. These locations were the wrist, waist and the shoe. Pedometers were excluded to guarantee the collection and processing of data on low intensity PA, where pedometers have failed (Cocker et al., 2012). Movements consisted of walking, running, and jumping. Each movement type was observed at different intensity levels (Slow 1m/s, Fast 2 m/s, Low/High intensity jump). 18.

(25) Furthermore sensor selection and placement were major issues that were addressed. The monitoring system identified the signal characteristics of different movement types and intensities on a complex laboratory based platform. This study also investigated the analysis methods of sensor signals, including the calculation of parameters such as the energy expenditure. To develop the collection of integrated information and to record exercise for long-term health management was an essential part of this study. Furthermore the purpose was to introduce a new theory of applied sensor technology in health and sport science in order to offer comprehensive and smart solutions to improve people’s health and sports skills.. 19.

(26) CHAPTER II. Method Participants. Fifteen healthy male university students with regular treadmill experience and without a record of any lower extremity injuries in the past 6 months volunteered to participate in the experiment. Participants were members of the Department of Physical Education of the National Taiwan Normal University.. Equipment. Equipment varied by each test. The measurement of the maximum jump height and the Preferred Transition Speed (PTS) tests were included in the Pre-test. The Main-test involved the analysis of different intensity and type of movements. Experiments were performed and recorded in the biomechanics laboratory of the Department of Physical Education of the National Taiwan Normal University. The Pretest was concluded in order to determine the PTS and the maximal jump height of each participant. Observations in Main-test demanded high-tech sophisticated devices as listed below. In both the Pre-test and the Main-test the Treadmill (Funa-7310, Tonic fitness technology, TW) was applied to control the speed of body movements (Fig. 1.). 20.

(27) FIGURE 1 Treadmill. In the Main-Test the Inertial Measurement Unit, (IMU-6000, Atomax, TW) with the Curo Mini software was applied to capture participants’ locomotion (Fig. 2.). The IMU incorporated a tri-axial accelerometer and gyroscope. This sensor fusion functioned as the IMU of the study. The IMUs were located on the shoe, waist and wrist. The range of acceleration change in all the three axes, three plants, the peak angular velocity and the resultant was recorded with the accelerometer. The peak angular velocity and the patterns of angular velocity on three axes were captured by the gyroscope.. 21.

(28) FIGURE 2 The IMU sensor. The. IMU. was. connected. with. a. cell. phone,. which. had. the. AMXExerciseAnalytics software specifically designed to capture IMU data as shown in Fig. 3.. FIGURE 3 The Software of the IMU 22.

(29) Protocol. All participants (n=15) were provided a familiarization period of few minutes to understand the equipment, the procedure and the laboratory environment and to get to know each personal in the experiment. After the above period participants were provided with the written informed consent forms. The study was proposed for ethical approval to the Institutional Review Board of the Taipei Medical University. Equipment was calibrated and set up. Participants were instructed to warm up thoroughly to avoid any injuries. Participants were tested individually. Altogether there were two experiments the Pre-test and the Main-test. The Pretest was concluded at least one day prior to the Main-test with each participant. Protocol of Pre-test: 1. Participants familiarize with the laboratory and its personnel 2. Participants are briefed about the entire procedure, and the treadmill 3. Sign Forms of Consent 4. Set up the equipment 5. Participants warm up 6. The height of the maximum power jump is measured 7. The individual Preferred Transition Speed, (PTS) of each participant is measured 8. Data is recorded and analyzed When the High intensity jump was defined there were three attempts of the maximum power jump. The highest point reached by participants was recorded and the lower results were not used in the study. The PTS measurement took 15 minutes per participant following the warm up. The experiment took place on the treadmill with four delta transducers attached to it. To avoid participants shifting into running or walking because of being aware of the transition speed the treadmill speed was hidden 23.

(30) from participants throughout the PTS test. It was especially important because most of the participants already had prior treadmill experience. First, participants walked at slow speed. The starting speed was set to 5 km/h, which is below the ever recorded PTS. The speed was increased by 0.2 km/h every 15 s until participants naturally shifted from walking into running. The start of the flight off phase indicated the moment of gait change when participants shifted into running. Running speed was increased to 8 km/h which is higher than average PTS but not higher than the Energetically Optimal Transition Speed (EOTS). Eight km/h was chosen as the top speed to avoid participants get fatigue. From this speed participants could slow down by reducing treadmill speed by 0.2 km/h every 15 s until it felt easier to walk. Participants were clearly instructed not to push walking. Participants could shift into running whenever it felt comfortable. When participants started to run the speed was recorded as Walk-Run PTS. The same instructions referred to Run-Walk transition. When participants decided to walk it was recorded as Run-Walk PTS. PTS was calculated by averaging Walk-Run (W-R) and Run-Walk (R-W) speeds. [(W-R) + (R-W)] /2 = PTS The order of Walk-Run and Run-Walk tests were switched at each participant to avoid the information exchange between them thus to avoid the learning effect. Protocol of the Main-test (Fig .4.): 1. Participants familiarize with the laboratory and its personnel 2. Participants are briefed about the entire procedure, and equipment 3. Sign Forms of Consent 4. Set up the equipment 5. Participants warm up 6. Sensors are positioned 7. Participants perform different intensity walk and run * 8. Participants’ jump with high intensity and jump with low intensity* 24.

(31) 9. Participants perform the Ten time low and the Ten time high intensity jump* 10. Data is recorded and analyzed *Steps 7-9. were performed in a randomized order. FIGURE 4 Protocol of the Main-test. Main-Test took approximately 2 hours per participant. The 3 IMUs were fixed on the side of the shoe, the pelvis (Lumbar vertebrae L3-4) and the wrist of participants. To ensure IMU placement and avoid movement of IMU positions an elasticated bandage was wrapped around the waist and the wrist. The IMU on the shoe was also fixed with high quality tape, which is for the purpose of supporting the position of reflective markers on the shoe. Locations of combined sensors are shown in Fig. 5.. 25.

(32) FIGURE 5 Wrist, pelvis, and shoe locations of combined sensors. After finishing the warm up, various intensity, and duration movements were measured. There were three types of movements. The Main-test encompassed different type, intensity and the same duration of movements. The 3 types of body movements were the Walk, Run and Jump (Fig. 6.).. 26.

(33) Slow(1m/s). Walk. Slow(1m/s) Run Fast(2m/s). Fast(2m/s). Low intensity Jump High intensity FIGURE 6 Three types of different body movements and two intensities. In the Main-test beyond the warm up procedure there was no accommodation time to adapt to the treadmill. Participants have all had a prior regular treadmill experience. Participants could draw which type of movement to start with. The 3 different types of locomotion were executed in randomized order. The purpose of the random order was to avoid the learning and the fatigue effects. Sufficient rest time was provided between the different types and intensity motions to further ensure that participants do not get fatigue. The Walk, and Run locomotion were executed with 2 intensities. The Jump movements were also executed with 2 intensities. There were Slow walk at the speed of 1m/s, Fast walk 2 m/s (above participants’ PTS) on the treadmill. The Run tests were also recorded at the same speed as the Walk, with Slow run at 1 m/s (below participants’ PTS) speed and Fast run at 2 m/s speed on treadmill. Participants were requested to perform each intensity level of Walk and Run movement for the duration of 2 minutes. The jump movement was performed with two intensities. Participants jumped with high intensity and then with low intensity. The low intensity jump was defined as the 30% of the height reached at the highest 27.

(34) maximum power jump of the Pre-test. The height of the Low intensity jump was calculated from two figures. The first height figure was the point where participants’ hands could reach up with arms completely vertically stretched. The second height figure was the highest point participants could reach with both hands after a High intensity jump. The first height figure was deducted from the second and the 30% of this height was defined as the Low intensity jump height. Participants had to reach this height for completing a Low intensity jump. The height was controlled with a ball hanging down from the ceiling of the lab. Participants were requested to perform each intensity jump three times.. Data Processing. The METLAB R2007b software (The MathWorks, MA, USA) was used to process the big data recorded during the tests. The 10 Hz low-pass filter was applied to reduce the noise. This four order Butterworth filtering process is similar to the method applied by Shih et al. (2014). The 10 Hz low-pass filtering was applied on the IMU to the gyro and accelerometer data. All noise was filtered except for the movement. The vibration of the IMU, especially on the intelligent shoe caused high frequency noise, which also had to be filtered. The force plate data was filtered by the 30 Hz low-pass four order Butterworth filtering technology. Five consecutive gait cycles of Run and Walk movements were analyzed. In order to gain reliable data three gait cycles would have been sufficient (Winter, 1984). Each phase was defined between the heel strikes of the left foot. The beginning and the end were marked by the peak points of deceleration on the anteroposterior axis due to the heel strikes. These two deceleration peaks determined one complete gait cycle. Other studies also defined the gait phase by measuring the acceleration data of the anteroposterior axis (Lee 2010, Zijlstra 2003). The intensity of the Walk and Run locomotion was defined by the range of acceleration displacement in between the two peak points. The lowest figures of acceleration data were also collected between the heel strike and toe off in the stance 28.

(35) phase. The mean value of acceleration data was calculated. These figures enabled the assessment of intensity. The range of acceleration change in three axes, and the peak angular velocity, from all different movement was calculated. The patterns of angular velocity were observed in order to find out the differences between the movements. The range of acceleration on all three axes, on three planes, and the resultant acceleration in each combination of the two axes (XY=horizontal plane, XZ=sagittal plane, and YZ=coronal plane, and all axes XYZ) were calculated. The range of acceleration was calculated using 3 axes and the Pythagorean theorem. Furthermore the gyroscope of the IMU recorded the maximum and minimum angular velocity in each axis and the graphic characteristics of the velocity curve variation were applied to benchmark data in order to find out the distinct characteristics of different intensity locomotion.. Statistical Method. The three-way repeated measures ANOVA was applied as the statistical method to determine the difference among movements, intensities and the locations of sensors. The significant level was set at α=.05. There were 3 movements, 2 intensities, and 3 sensor positions. Therefore the three-way ANOVA (3 movements × 2 intensities × 3 sensors) with repeated measures on test variables was the statistical method to analyze all data. At significant main effects, Bonferroni post hoc tests were applied to identify the statistically significant mean differences. The level of significance was set at α =. 05. When the interactions among movement, intensity and sensor positions were all significant, then the movement and intensity or positions were separately measured and only two factors were compared with each other.. 29.

(36) CHAPTER III. Results In the present study the Three-way ANOVA (3 movements × 2 intensities × 3 sensors) with repeated measures on all test variables was applied. The statistical and mathematical approach demonstrated significant results. There was a significant interaction effect among the movement, intensity, and the sensor positions. Most of the results of the different locomotion varied at low and high intensity levels and by IMU locations as well. After the three-way ANOVA a simple effects test was performed. The interaction between only the intensities and the movements provided nonsignificant results. The waist IMU measured the least significant interaction among different movements and intensities. Both the shoe and the waist IMU regarding the acceleration captured non-significant interaction between the intensities and movements. There was no significant interaction on the waist IMU’s superior–inferior axis (A-z G). Furthermore the results on the A-yz (F=2.68, p=0.11) and the G-x (F=0.008, p=0.99) axis picked up by the wrist IMU; on the A-y (F=1.5, p=0.24), A-z (F=0.95, p=0.38), A-yz (F=2.99, p=0.09), A-xyz (F=5.02, p=0.14), G-z (F=0.43, p=0.66) axis picked up by the waist IMU; on the A-y (F=2.06, p=0.17), A-xz(F=1.48, p=0.25), A-xyz (F=1.62, p=0.22) axis picked up by the shoe IMU trended to show no significant interaction effects between the intensity and movement (Table1.).. 30.

(37) Table 1. Results of interaction effects measured by the 3 IMUs Wrist Low intensity High intensity Walk Run Jump Walk Run Jump A-x(G) 0.65±0.12*bc 1.81±0.60*ac 8.38±3.43*ab 1.16±0.16bc 2.67±0.43ac 16.01±2.29ab A-y(G) 0.41±0.14*bc 1.02±0.41ac 3.99±0.95*ab 0.90±0.37c 1.19±0.37c 5.94±1.47ab A-z(G) 0.36±0.16*bc 3.14±0.80*ac 5.03±0.99*ab 1.33±0.36bc 4.39±0.79ac 7.46±1.82ab A-xy(G) 0.70±0.11*bc 1.52±0.60*ac 7.65±2.90*ab 1.13±0.16bc 1.96±0.35ac 15.23±1.99ab A-xz(G) 0.66±0.12*bc 3.03±0.68*ac 8.04±2.95*ab 1.26±0.19 bc 3.91±0.70 ac 15.68±2.26 ab A-yz(G) 0.46±0.16 2.96±0.67 4.62±0.87 1.13±0.30 3.75±0.70 6.05±1.67 A-xyz(G) 0.71±0.10*bc 3.11±0.71*ac 8.39±2.73*ab 1.28±0.21bc 3.97±0.71ac 15.96±2.03ab G-x(d/s) 188.07±57.85 246.98±45.94 847.20±176.24 269.87±74.46 327.70±133.91 934.16±239.40 G-y(d/s) 237.31±53.78*bc 361.46±101.22*ac 1277.13±305.59*ab 524.03±175.13c 450.19±134.44c 1630.89±277.49ab G-z(d/s) 112.26±55.58*bc 180.61±70.86*ac 718.67±199.40*ab 176.92±86.90bc 272.24±48.21ac 942.73±181.42ab Waist Low intensity High intensity Walk Run Jump Walk Run Jump A-x(G) 0.54±0.07*c 0.48±0.09*c 1.40±0.21*ab 1.19±0.18bc 0.70±0.11ac 3.04±1.28bc A-y(G) 0.49±0.09 0.62±0.15 0.83±0.34 0.89±0.20 0.87±0.19 1.18±0.31 A-z(G) 0.62±0.19 2.02±0.44 4.18±0.68 1.40±0.15 2.71±0.42 5.12±0.78 A-xy(G) 0.41±0.06 0.52±0.11 1.05±0.21 0.94±0.14 0.67±0.10 2.24±0.78 A-xz(G) 0.64±0.19*bc 1.93±0.29*ac 3.73±0.55*bc 1.54±0.14 bc 2.39±0.26 ac 4.43±0.73 ab A-yz(G) 0.64±0.19 1.93±0.29 3.72±0.56 1.44±0.16 2.39±0.26 4.30±0.75 A-xyz(G) 0.67±0.18 1.94±0.29 3.75±0.55 1.58±0.16 2.44±0.28 4.45±0.73 G-x(d/s) 53.46±17.35* 66.54±20.36* 56.31±18.83* 105.49±16.22bc 83.87±23.68a 85.93±18.93a G-y(d/s) 43.33±8.78*bc 62.89±17.20*ac 206.00±32.85*ab 70.95±16.82bc 87.20±24.27ac 347.26±57.67ab G-z(d/s) 72.81±19.35 87.70±21.22 85.02±24.67 136.92±41.70 155.09±38.61 136.95±47.57 Foot Low intensity High intensity Walk Run Jump Walk Run Jump A-x(G) 4.57±0.76* 4.83±0.75* 4.56±0.81* 8.03±0.82b 6.39±0.88ac 8.57±1.08b A-y(G) 1.57±0.39 1.57±0.51 2.24±0.78 2.45±0.44 2.20±0.42 3.48±1.10 A-z(G) 2.45±0.35*c 2.43±0.47*c 5.33±1.04ab 3.67±0.38bc 2.76±0.68ac 5.60±1.05ab A-xy(G) 3.78±0.63* 4.08±0.54*c 3.12±0.58*b 6.16±0.45b 5.60±0.68a 5.63±0.77 A-xz(G) 4.33±0.76 4.67±0.63 4.36±0.53 6.34±0.54 6.14±0.86 5.91±0.72 A-yz(G) 2.44±0.32*c 2.49±0.48*c 4.27±0.60ab 3.67±0.34 bc 3.03±0.61 ac 4.60±0.78 ab A-xyz(G) 4.50±0.74 4.81±0.64 4.47±0.64 6.42±0.50 6.11±0.77 6.05±0.69 G-x(d/s) 320.39±74.75* 385.29±58.56* 356.01±100.63* 435.86±92.25bc 561.74±71.37a 670.34±219.46a G-y(d/s) 765.96±57.48*c 680.19±82.30*c 868.55±135.48ab 1105.53±66.99bc 808.54±53.80ac 885.16±68.00ab G-z(d/s) 237.75±54.14* 213.89±47.51* 251.66±100.29* 280.16±42.35c 271.92±88.63c 465.75±175.75ab # significant interaction effect; *significant difference between intensity levels a denotes the significant differences in walking; b denotes significant differences in running; and c denotes significant differences in jumping. 31. F 73.062 167.74 8.843 84.929 89.727 2.676 93.602 0.008 9.161 8.714. P <.001# <.001# 0.005# <.001# <.001# 0.114 <.001# 0.992 .001# .008#. F 14.904 1.50 0.953 22.616 6.332 2.988 5.022 13.886 33.891 0.429. P <.001# 0.24 0.382 <.001# 0.005# 0.088 0.14 <.001# <.001# 0.655. F 22.80 2.056 5.763 6.712 1.478 5.237 1.617 8.093 41.425 17.169. P <.001# 0.169 0.008# 0.004# 0.245 0.024# 0.217 0.006# <.001# <.001#.

(38) During the statistical analysis the main effects of intensity and movement showed significant difference. The main effects analysis of the acceleration of the anteroposterior direction provided significant differences among the three movements (F=59.08, p < .01). At Walk, Run, and Jump movements the same trends of acceleration change could be observed at different intensities. As the movement intensity increased the values of all parameters increased. Run and Walk patterns consistently exhibited the similar increasing trend that was picked up by the shoe IMU. The angular velocity captured by the shoe IMU in the anteroposterior X axis, and the mediolateral Y axis produced significant data. However at the Jump movement the angular velocity on the vertical Z axis did not produce significant results (F=0.64, p=0.44). At the same time the angular velocity of both the Walk and Run movements grew significantly by the increase of intensity, which produced significant results on the vertical Z axis. When the different intensity movements were compared by acceleration significant differences trended to appear in almost all axes. However there was no significant displacement on the vertical axis either at Low Intensity Jump or at High Intensity Jump. This may have been caused due to the plantar-flexion movement during takeoff and landing. The peak angular velocity captured by the shoe IMU similarly to the acceleration results at both the High Intensity Jump and the Low Intensity Jump did not result in significant displacement either at plantar-flexion or dorsiflexion. The post hoc tests on the angular velocity of the plantar and dorsiflexion in the frontal axis (Y axis) resulted in consistent differences among Walk, Run, and Jump movements that could identify different intensities. However the trends of movement modes varied (Fig.7.).. 32.

(39) FIGURE 7 The result of acceleration (left) and angular velocity (right) in X and Y axis (shoe) respectively * significant difference among different movements.. As the intensity increased, the acceleration values of all parameters proportionally increased, especially at Run movement. These results were caught up by the shoe IMU. The values at Run were greater than at Walk or Jump. Both the multi-axial and the uniaxial parameters were able to identify the different intensities of movements. However the multi-axial parameters were not as effective as the uniaxial parameters in determining the intensity of locomotion. Cut points to separate the different intensities of movements were set up based on velocities captured by the shoe IMU. The most significant increase at both the acceleration on the X axis, and the angular velocity on the Y axis were observed at the Run movement. The gait cycle was determined by acceleration data. The time between the heel strikes of the left foot defined the epoch or the shortest observation period of each body movement. All calculations were done with the comparison of these cycles. Each peak on the anteroposterior axis was caused by the heel contact. Patterns of angular velocity could also identify one cycle on the Y axis. The first peak appeared at the plantar-flexion between the heel contact and the toe contact. The second peak could be observed from the mid-stance phase to toeoff marked with the line in the middle. In between these two peaks a flat curve trended to show up. The value was very close to zero. The combined results of accelerometer and gyro clearly identified gait cycles of walking (Fig. 8.). 33.

(40) Heel contact Heel contact. Swing phase. Stance phase. FIGURE 8 Definition of one gait cycle at Slow walk from the shoe IMU. In the Walk movement patterns the acceleration of anteroposterior axis and the angular velocity of frontal axis were more representative than others due to the relatively high values and the regular changing of walk patterns. A regular sequence was measured in the occurrence of plantar and dorsiflexion (Fig. 9.).. FIGURE 9 Data of Slow walk from six axes (gyro on the left, accelerometer on the right) 34.

(41) In the comparison of different intensity walk the peak values of the Fast walk were significantly higher than the Slow walk. Significant differences were observed between the peak to peak intervals. The intervals of Fast walk were significantly shorter than the intervals of Slow walk measured by the shoe IMU (Fig. 10.). Slow Walk. Fast Walk. *. *. FIGURE 10 Comparison of Slow walk and Fast walk intervals * Significantly shorter intervals. In the comparison of Slow Run and the Fast Run movement results showed a similar trend as between the Slow and the Fast walk. However the peak values were significantly higher and the intervals were shorter at Run movement measured by the shoe IMU (Fig. 11.).. 35.

(42) Slow Run. Fast Run. *. FIGURE 11 Comparison of the Slow run and the Fast run * Significantly shorter interval. At Slow walk the acceleration data of the anteroposterior axis and the gyroscope data of the foot rotation on Y-axis exhibited a regular sequence. The two patterns identified the same heel contact time at the low intensity Slow walk caught up by the shoe IMU (Fig.12.).. FIGURE 12 Shoe IMU data at the Slow walk. 36.

(43) Similarly to the phenomenon at Slow walk the accelerometer on the anteroposterior axis and the Gyroscope on the Y-axis of the shoe IMU at Slow run trended to have an agreement. Both clearly showed the heel strike at the same time demonstrating it with the highest range of displacement between the high peak and the low peak as shown in Fig 13.. FIGURE 13 Shoe IMU data at Slow run. At Fast walk the acceleration on the anteroposterior axis and the gyroscope data on the superior-inferior axis trended to show a regular sequence of movement cycles (Fig.14.).. FIGURE 14 Waist IMU data at Fast walk 37.

(44) The waist IMU at Fast Run did not trend to deliver consistent sequence of gait cycles and the displacement between the positive and negative peaks did not show significant results either. At Fast run movement neither the accelerometer nor the gyroscope discovered matching patterns (Fig. 15.).. FIGURE 15 The waist IMU patterns at Fast run. The IMU fixed on the wrist at Slow run provided a clear pattern, which demonstrates the consistent waving movement of participants arm. This phenomenon was captured on the Y axis of the gyroscope and on the X axis and the Z-axis of the accelerometer (Fig.16).. FIGURE 16 The wrist IMU patterns at Slow run 38.